Patients Like Mine notebook

Patients Like Mine is intended as a tool to help search for patients that are highly similar to a corresponding patient of interest using clinical events data. You initially choose a patient of interest and then use this tool to identify similar matching patients. This feature computes a similarity score for each patient based on an algorithm called Weighted Personalized PageRank (PPR).

For more information, see:

• Underlying concepts
  
• What it looks like
  

Underlying concepts

For an intuitive understanding of Weighted PPR, we first explain two graph-related techniques, namely, Random Walk and PageRank.

• Random Walk: A random walk on a given graph is an iterative process starting from a random vertex and in each step, either follows a random outgoing edge of the current vertex or jumps to a random vertex (since some vertices might not have any outgoing edges).
• PageRank (PR): This algorithm computes a distribution of a specific kind of random walk that starts from a random vertex and in each iteration, with a pre-defined probability p, jumps to a random vertex, and with probability 1-p follows a random outgoing edge of the current vertex. Running the page rank algorithm on a graph will generate rankings (PR value) for vertices and the numeric PR values can be viewed as "global importance" of vertices.
• Personalized Page Rank (PPR): This algorithm is an extension of the classical PR such that jumps are back to one of a given set of starting vertices. In a way, the walk in PPR is biased (or "personalized") towards this set of starting vertices and is more localized compared to the random walk performed in classical PR.

Lastly, the weighted variant of PPR (used in this specific notebook) assigns weights on edge types to highlight the relative importance of different edge types while executing the random walks.

What it looks like

The Patients Like Mine displays the graphical view of similar patients having a diagnosis and procedures done.

Figure 2-1 Healthcare Graph

The Patients Like Mine notebook also provides the user with the capability to adjust the weights of the clinical attributes based on their interest which would enable them to personalize the similarity search feature to their use case as shown below.

Once the weights are set and algorithm is executed, you would see a list of patients ordered with higher PPR scores as shown below.

The highest scored patients are most similar matching patients to our patient of interest. Now you can drill down further by selecting one of the matching patients and visualize the one-to-one graph with similar clinical attributes between these two patients as shown below.

Finally, you can also look at which clinical attributes are matching between these two patients and which clinical attributes are different as shown below.

We also added the Kaplan-Meier estimator for comparing multiple procedures in terms of the survival probability either across similar patients or across all patients as shown below.

Figure 2-2 Kaplan-Meier Analysis of Differently Treated Cohorts

Figure 2-3 Kaplan-Meier Estimator